Re: [boost] build_monitor.exe: a cure worse than the disease

From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2004-08-02 05:09:16

Victor A. Wagner Jr. writes:
> At Sunday 2004-08-01 16:37, you wrote:
> >Victor A. Wagner Jr. writes:
> > > I have NO idea what criteria build_monitor uses to killing
> > > processes,
> >
> >Two conditions:
> >
> >1) They have to be spawned by 'bjam.exe' executable.
> 
>          how do you know?

By recursively obtaining their parent processes using 
'NtQueryInformationProcess'
(http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dllproc/base/ntqueryinformationprocess.asp) 
and checking if any of them is named 'bjam.exe'.

> 
> >2) They have to be running for more than 5 minutes or show an
> >    error dialog window.
> 
>          these other things run 100% of the time on my system. in the 
> middle of the night they may even go for 5 minutes with NO I/O.

In case I wasn't clear, *both* of the above conditions have to hold for 
the process to become a candidate for termination.

> 
> > > but when I awoke this morning, it had managed to kill:  the dnet
> > > "run in the background" ORG process, Pirch (my IRC client), a
> > > proxy that lets me get dcc's through my hardware firewall,
> > > Trillian (the instant message program), SpyBot resident.....
> > > I think you get the picture.
> >
> >How did you determine that it's build_monitor that caused all
> >that? While there is always a possibility of a program having bugs,
> >it's hard to believe that this particular one started to misbehave
> >all of a sudden after a year of stable work and no changes to the
> >source code.
> >
> > > this thing just walked through my system like a hoard of Vandals
> > > (or was it Visigoths). whatever
> >
> >I'm sorry to hear about the whole thing, but let's get to the roots
> >of it. First of all, why do you think it was build_monitor? Second,
> >what operating system are you running?
> 
> 1) My system has run flawlessly for over a year also and two days ago was 
> when I first installed monitoring.....
> the first time it ran a bunch of stuff wasn't working in the morning, but 
> I'd been very tired when I went to bed and figured MAYBE I'd shut them down 
> (unlikely as I _never_ _NEVER_ shut down the dnet client)...but I let it 
> go... I made some changes to regression.py so it would work better here 
> then manually ran the system that afternoon...no problems.  I let it run 
> automatically again overnight and "bam" all these things that are ONLY 
> started at login time and NEVER touched by me are missing again....
> the _logical_ place to look is at "what's new?"... what's new 
> is  -monitored on the test line in my script.

Understood. Well, there is a slight chance that the layout of 
PROCESS_BASIC_INFORMATION structure on WindowsXP is different from the one
on Windows 2000 (the stucture is "for internal use"), and that could cause 
the kind of reckless behavior you've seen.

> btw, this morning I also noticed that build_monitor had it's own window 
> open and was spewing out TONS or data to it... is that normal?

Yep. It simply logs everything it does. 

> 
> I took out the teststreams test (I think) but this afternoon's run had 
> already started... but I chose ignore when the modal dialog (may the .....) 
> popped up.  btw, tho my log clearly shows them sent to the ftp site, 
> they're not on the website yet... the date is Sunday the time 21:30:03 in 
> the xml file

There are there now.

> 
> I'd be more than happy to help find this as any problems in running the 
> regression tests will just mean less folks willing to run them.  

Thank you.

> if you get 
> back to me quickly, I can turn on the "-monitored" again... the run starts 
> in just over 2 hours (on the 1/2 hour) (still missing teststreams at this 
> point).

Let us test on Windows XP locally first. 

> 
> 2) WinXPpro sp1

Thanks!

--
Aleksey Gurtovoy
MetaCommunications Engineering